Conversation
execute_block_parallel previously returned (current_state, codes, shadow_touched, shadow_reads) per tx and validated them in a serial post-loop across all txs (validate_tx_execution + shadow checks + mark unread_storage_reads / unaccessed_pure_accounts). Validation is per-tx pure work; only marking the shared sets mutates cross-tx state. Move validate_tx_execution + shadow_touched/shadow_reads checks into the closure. Precompute the small (Vec<(Address, H256)>, Vec<Address>) inputs the serial pass needs to update the shared sets. Drop current_state + codes inside the closure so they no longer cross the rayon boundary. Defer BAL validation errors via Option<EvmError> in the result tuple so the post-par_iter gas-limit check still takes priority over BAL mismatch (preserves GAS_USED_OVERFLOW > BAL error ordering for blocks exceeding the gas limit). Expected win on 200-tx blocks: the ~3 ms median serial validation pass goes away. Also reduces per-tx allocator pressure across rayon workers since the per-tx maps are dropped before the result tuple is constructed.
- reads_satisfied: pre-size with current_state.len() * 4 to skip 2-3 reallocations on the hot path (rough avg slots-per-account). - destroyed: keep Vec::new() since selfdestruct is rare post-EIP-6780; no-allocation default is optimal. - Document why the post-collect sort_unstable_by_key is a defensive no-op (IndexedParallelIterator preserves order) so a future refactor doesn't drop the guard. Addresses ElFantasma feedback on #6677.
Replace N parallel point-gets in CachingDatabase::prefetch_accounts with a single multi_get_cf on ACCOUNT_FLATKEYVALUE for FKV-covered addresses, falling back to per-address trie walks for paths past the FKV cursor. Reduces warmer-phase cost on the bal-devnet-7 fixture from 6.78 ms to 2.37 ms per block (-65%); total wall time -21.5%.
Address review feedback on PR #6712: - store.rs: hoist `fkv_cursor` out of the per-address loop and document why the comparison matches `BackendTrieDB::flatkeyvalue_computed`'s semantics (not the more-conservative `flatkeyvalue_computed_with_last_written`). Drop the dead `diff_hits` counter. - vm.rs: comment on why `cache.insert` (vs `or_insert`) is intentional. - rocksdb.rs: clone a `String` instead of formatting the error per key.
Address review feedback on #6712: - Parallelize the trie-walk fallback in get_account_states_batch_by_root with rayon par_iter, recovering the per-address fan-out that the pre-batch prefetch_accounts path had. Matters during initial sync when most addresses miss the FKV cursor; no effect on the FKV-hot path. - Drop per-iteration Nibbles::from_hex(path.clone()) in the classifier loop; compare last_written and the leaf path as slices directly (Nibbles::Ord delegates to data Vec compare, so this is equivalent). - Expand multi_get rustdoc to spell out that the default impl is equivalent to N get calls on backends without a batched primitive.
…96) for glamsterdam-devnet-4 Extends PayloadAttributesV4 with the new required target_gas_limit field introduced by execution-apis#796 and consensus-specs#5235 (Gloas / upstream "Glamsterdam"). The CL is now the authoritative source of the per-block target gas limit on engine_forkchoiceUpdatedV4 instead of the EL reading it from the static --builder.gas-limit flag. validate_attributes_v4 strictly rejects FCUv4 requests that omit the field with RpcErr::InvalidPayloadAttributes, matching the spec contract. The paired CL on glamsterdam-devnet-4 ships consensus-specs#5235 and populates the field on every FCUv4 it sends. build_payload_v4 sets BuildPayloadArgs.gas_ceil directly from the attribute. The EIP-1559 1/1024 gas-limit envelope inside calc_gas_limit continues to clamp the value relative to the parent block. V1/V2/V3 and non-engine callers (sequencer, dev mode) keep using --builder.gas-limit. Also fixes a latent collision in BuildPayloadArgs::id(): slot_number was never hashed and gas_ceil now matters per-call, so two FCUv4s differing only in CL-supplied targetGasLimit could collide on payload_id. Both fields are now part of the Keccak input. Tests added: - payload-attribute parser round-trips for hex value present, field absent, and explicit null - validator accepts present, rejects absent (matches spec error message), and continues to reject pre-Amsterdam timestamps - BuildPayloadArgs::id() distinguishes different gas_ceil values and different slot_number values, and is stable under equal inputs - calc_gas_limit steps up/down toward target within one EIP-1559 1/1024 step and clamps when the target is within one step
- Combine the per-protocol "Negotatied eth/snap version" debug pair into a single structured line that includes the peer node and client id. Two unrelated-looking lines for the same connection are now one. - Fix the `Negotatied` typo in the same site. - Demote `early eof` IO errors during the handshake to TRACE; they are expected on simultaneous-dial races (both peers initiate, only one connection wins). Other handshake failures stay at DEBUG with structured `error` field.
|
Lines of code reportTotal lines added: Detailed view |
Benchmark Results ComparisonNo significant difference was registered for any benchmark run. Detailed ResultsBenchmark Results: BubbleSort
Benchmark Results: ERC20Approval
Benchmark Results: ERC20Mint
Benchmark Results: ERC20Transfer
Benchmark Results: Factorial
Benchmark Results: FactorialRecursive
Benchmark Results: Fibonacci
Benchmark Results: FibonacciRecursive
Benchmark Results: ManyHashes
Benchmark Results: MstoreBench
Benchmark Results: Push
Benchmark Results: SstoreBench_no_opt
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
[DNM] — Do Not Merge. Opened to run CI on the rebased
glamsterdam-devnet-4branch.This branch was rebased onto
mainvia cherry-pick. 5 commits already present onmainwere dropped (FCU finalized-ancestor work via #6676; BAL storage prefetch via #6732). The 10 commits below are the net-new work:multi_get_cf(perf(l1): batch account-state prefetch via rocksdb multi_get_cf #6712)get_account_state_multitargetGasLimit(execution-apis#796)Not for merge; tracking CI only.